define-brand-guidelines
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from
CUSTOMER.mdto drive its core logic and generate recommendations, which introduces a surface for indirect prompt injection if the source file contains malicious instructions. - Ingestion points: The skill reads
CUSTOMER.mdin Phase 0 and Phase 1 to extract ICP definitions, pain points, and values to pre-fill user questions. - Boundary markers: The instructions do not define delimiters (e.g., XML tags or triple quotes) or provide specific 'ignore' instructions for the agent when processing the content of
CUSTOMER.md. - Capability inventory: The skill uses
globfor file discovery,AskUserQuestionfor user interaction, and possesses file-writing capabilities to generateBRAND_GUIDELINES.md. - Sanitization: There is no validation or sanitization logic to verify the integrity or safety of the content read from the local filesystem before it is used to influence agent behavior.
Audit Metadata